package mlbdd

  1. Overview
  2. Docs

OCaml BDD

A library for reduced ordered binary decision diagrams (ROBDDs) written in OCaml

Introduction

This is a simple library that implements reasonably efficient binary decision diagrams in OCaml. A binary decision diagram (BDD) is a canonical representation of a boolean formula. It represents the standard elements including and, or, not, variables, and the constants true and false.

This is intended to serve as a nice middle ground between very primitive BDD implementations written in OCaml, and OCaml wrappers around high-performance C/C++ libraries. What differentiates this implementation from other simple implementations is the tracking of complement edges. This means that the not operation (dnot) has O(1) complexity, as opposed to O(n) complexity, where n is the number of nodes in the representation of the BDD (potentially exponential with respect to the number of variables).

The API of this library is styled after mlcuddidl, which is a wrapper around the high-performance CUDD library. The mlcuddidl library is available here.

This implementation of BDDs is developed and maintained by Arlen Cox and is made available under the MIT license.

Example

All of the operations on BDDs are performed with respect to a manager. A manager is responsible for keeping track of tables of all of the allocated BDD nodes across all of the BDDs created from that manager. BDDs from two different managers must never be mixed.

To create a manager, the init function below can be used. The following code constructs a manager man and constructs a BDD representing the 0th variable.

let man = init ~cache:1024 () in
let v0 = ithvar man 0 in

Variables, are represented by integers. Lower numbers are closer to the root of the decision tree and higher numbers are closer to the leaves of the decision tree. We have already seen the variable 0 constructed above, but constructing the variable 0 again (even if we bind it to a different OCaml name) is equal to the original variable 0:

let v1 = ithvar man 0 in
assert(equal v0 v1)

This is because BDDs are canonical. Two formulas that are logically equivalent have a single representation in the BDD, no matter how they are constructed.

Formulas can be constructed by calling constructor functions such dtrue, which constructs the formula representing true. Similarly, the formula that represents the conjunction of two variables a and b can be constructed in the following way:

let a = ithvar man 0 in
let b = ithvar man 1 in
let ab = dand a b in

Furthermore, it is possible to constrain that three variables are equal to each other:

let a_eq_b = eq a b in
let c = ithvar man 2 in
let b_eq_c = eq b c in
let eq_abc = dand a_eq_b b_eq_c in

Because this is a canonical representation, it wouldn't matter if b_eq_c were constructed with eq c b, eq a c, or eq c a. The result would always be the same.

let eq_abc2 = dand a_eq_b (eq c b) in
assert(equal eq_abc eq_abc2);
let eq_abc3 = dand a_eq_b (eq a c) in
assert(equal eq_abc eq_abc3);
let eq_abc4 = dand a_eq_b (eq c a) in
assert(equal eq_abc eq_abc4);

Additionally, BDDs support quantification, so we can existentially quantify out b from eq_abc to get that a = c:

let eq_ac = exists (support b) eq_abc in
assert(equal eq_ac (eq a c))

Finally, there are some satisfiability and optimization options. For example, we can determine that eq_ac is satisfiable.

assert(sat eq_ac <> None);

Additionally, it is possible to iterate over a number of satisfying assignments:

itersat (fun sat ->
      print_endline "-----------"
        List.iter (fun (is_pos,v) ->
            if not is_pos then
              print_string "~";
            print_endline (string_of_int v)
          ) sat
    ) eq_ac

Similarly, it is possible to iterate over and print out all of the prime implicants of the formula.

iterprime (fun sat ->
      print_endline "-----------"
        List.iter (fun (is_pos,v) ->
            if not is_pos then
              print_string "~";
            print_endline (string_of_int v)
          ) sat
    ) eq_ac

API Documentation

type var = int

Variables have type var, which is an integer

type man

man is the type of a BDD manager that is passed to primitive constructors ithvar, dtrue, and dfalse

type t

t is the type of a BDD. It is the root of a directed acyclic graph that represents the formula

type support

support is the type of support for a BDD. A support is the set of variables that are needed to represent a formula

type 'r hist

'r hist is a type that keeps track of fold history. This allows folding over multiple BDDs without revisiting nodes.

Manager Manipulation and Meta-Functionality

val init : ?cache:int -> unit -> man

init () creates a new BDD manager with a default cache size.

init ~cache:n () creates a new BDD manager with a cache of size n.

Larger caches require fewer growth steps, where the cache must be resized and memory must be reallocated.

val clear : man -> unit

clear man clears all of the caches in the BDD. BDDs are still accessible after clearing the cache, but canonicity may be lost. It is recommended that this command not be used unless necessary. It is likely much safer to allocate a new manager than to clear the cache of an existing one.

val flush : man -> unit

flush man clears the input caches in the BDD. BDDs are still accessible and extensible with full canonicity. This may decrease allocation performance because extra BDD nodes may be allocated and then immediately thrown away.

val manager : t -> man

manager t retrieves the manager for the BDD expression t.

val support : t -> support

support t computes the supporting variable set for the bdd t.

val list_of_support : support -> int list

list_of_support support converts a support structure support into a list of variable indexes that constitute the support

val support_of_list : int list -> support

support_of_list l converts a list of variables l into a support structure that can be used for quantification

val string_of_support : support -> string

string_of_support support creates a string representation of a supporting set.

Queries

val is_true : t -> bool

is_true t returns true if the BDD is the formula true.

val is_false : t -> bool

is_false t returns true if the BDD is the formula false.

val equal : t -> t -> bool

equal t1 t2 returns true if the two BDDs represent the same function.

val id : t -> int

id t returns a unique indentifier for a particular node. This can also be used as a hash code

val hash : t -> int

hash t returns a hash code for a particular node. Note hash codes should not be persisted from run to run.

Constructors

val dtrue : man -> t

dtrue man returns the BDD representing the formula true.

val dfalse : man -> t

dtrue man returns the BDD representing the formula false.

val ithvar : man -> var -> t

ithvar man v returns th BDD representing the variable v. If v is true, the formula is true, otherwise if v is false, the formula is false

val dnot : t -> t

dnot t returns the complement or negation of the BDD t.

val dand : t -> t -> t

dand t1 t2 returns the BDD representing the conjunction of t1 and t2.

val dor : t -> t -> t

dor t1 t2 returns the BDD representing the disjunction of t1 and t2.

val nand : t -> t -> t

dor t1 t2 returns the BDD representing the negation of the conjunction of t1 and t2.

val xor : t -> t -> t

dxor t1 t2 returns the BDD representing the exclusive-or of t1 and t2.

val nxor : t -> t -> t

nxor t1 t2 returns the BDD representing the negation of the exclusive-or of t1 and t2.

val eq : t -> t -> t

eq t1 t2 returns the BDD that is true whenever t1 is equal to t2 (equivalent to nxor)

val ite : t -> var -> t -> t

ite f v t returns the bdd that is f whenever the variable v is false and t whenever the variable v is true

val imply : t -> t -> t

imply t1 t2 returns the BDD that represents t1 implies t2

val exists : support -> t -> t

exists support t returns the BDD where each variable in support has been existentially quantified in t

val forall : support -> t -> t

forall support t returns the BDD where each variable in support has been universally quantified in t

val cofactor : var -> t -> t * t

cofactor v t returns the negative and positive cofactor of t with respect to variable v

val permute : var array -> t -> t

permute p t permutes the variables in the BDD t using the permutation array p. At each variable's index in p is the new index for that variable.

val permutef : (var -> var) -> t -> t

permute f t permutes the variables in the BDD t using the permutation function f. The function returns a new index for each index.

Iteration

type 'a e =
  1. | False
    (*

    the false BDD

    *)
  2. | True
    (*

    the true BDD

    *)
  3. | Not of 'a
    (*

    not result

    *)
  4. | If of 'a * var * 'a
    (*

    result from false branch, variable, result from true branch

    *)

type for accessing BDD nodes as they are used

val inspect : t -> t e

inspect t returns the top node as an iteration expression

type 'a b =
  1. | BFalse
    (*

    the false BDD

    *)
  2. | BTrue
    (*

    the true BDD

    *)
  3. | BIf of 'a * var * 'a
    (*

    result from false branch, variable, result from true branch

    *)

type for accessing BDD nodes as traditional BDDS

val inspectb : t -> t b

inspectb t returns the top node as an traditional BDD node

val fold : ('r e -> 'r) -> t -> 'r

fold f t folds over the BDD's structure calling f on each node in the BDD.

val foldb : ('r b -> 'r) -> t -> 'r

fold f t folds over the BDD's structure calling f on each node in the BDD. Follows the traditional BDD structure.

val fold_init : man -> 'r hist

fold_init man creates a new hold history tracker. Use this along with the fold_cont or foldb_cont operations to perform folds using a shared history.

val fold_cont : 'r hist -> ('r e -> 'r) -> t -> 'r

fold_cont hist f t continues a fold with an existing history hist. Use fold_init man to create a new history.

val foldb_cont : 'r hist -> ('r b -> 'r) -> t -> 'r

foldb_cont hist f t continues a fold with an existing history hist. Use fold_init man to create a new history. Fold uses the traditional BDD structure

Satisfiability and Primality Queries

val sat : t -> (bool * var) list option

sat t returns None if unsatisfiable, otherwise it returns a list of assignments and the corresponding variable that would make the formula true. Note that this may not be a complete satisfying assignment. Not every variable int the support is necessarily included.

val allsat : t -> (bool * var) list list

allsat t returns the complete list of satisfying assignments that could possibly be returned by sat.

val itersat : ((bool * var) list -> unit) -> t -> unit

itersat f t iterates over the complete list of satisfying assignments that could possibly be returned by sat, calling f on each one

val prime : t -> (bool * var) list option

prime t returns None if unsatisfiable, otherwise it returns a prime implicant of the formula t. A prime implicant is is represented as a list of polarities and variables. If the polarity is true, then the variable appears unnegated in the implicant. If the polarity is false, then the variable appears negated in the implicant.

val allprime : t -> (bool * var) list list

allprime t returns the list of all prime implicants of the formula t. A prime implicant is is represented as a list of polarities and variables. If the polarity is true, then the variable appears unnegated in the implicant. If the polarity is false, then the variable appears negated in the implicant.

val iterprime : ((bool * var) list -> unit) -> t -> unit

iterprime f t iterates over the complete list of prime implicants could possibly be returned by allprime, calling f on each one

String Conversion

val to_string : t -> string

to_string t returns a string representation of the BDD. It represents the internal structure the following way: ~ represents negation of a variable for a term. F represents the false terminal. T represents the true terminal. An If-Then-Else structure is represented as a 4-tuple (var, if_false, if_true, id), where if_false is the BDD if the var is false and if_true is the BDD if the var is true. Id is the unique identifier for this BDD in the cache.

val to_stringb : t -> string

to_stringb t returns a similar string representation, but it removes complement edges, so that the BDD is shows as the formula would be represented in traditional If-Then-Else normal form

module Raw : sig ... end
module type WHS = sig ... end
module WeakHash (H : Hashtbl.HashedType) : WHS with type key = H.t
OCaml

Innovation. Community. Security.